Pruning Strategy on Adaptive Rule Model by Sorting Utility Items

Abstract

The adaptive Rule Model is an association rule development that formulates a minimum threshold value according to the data characteristics. The formulation process is based on item frequency and utility for other considerations, which requires high runtime and memory usage. For better performance, it is necessary to develop a pruning strategy that can minimize runtime and memory usage without reducing the quality of the rule produced. Therefore, this research aims to develop 2 pruning strategies, namely the Adaptive Rule with Average Utility (ADR-AU) and the Adaptive Rule with Sorting Utility (ADR-SU) to improve the adaptive rule model. The ADR-AU changes the minimum threshold using the average utility item. The ADR-SU model proposes an additional step in the frequent itemset determination process. It is used for sorting utility items in descending order before the iteration process is carried out to compare whether the values are greater than the minimum threshold. Based on the experiments on 8 datasets, the number of frequent itemset in the ADR-AU and ADR-SU models was significantly less in the adaptive rule model. The results showed that all datasets experienced a decrease in runtime between the adaptive rule model and ADR-AU. Meanwhile, for the Zoo and Sco datasets, there was an increase in runtime between the ADR-AU and ADR-SU models. In terms of memory usage, the connect and retail datasets experienced a decrease, while the other 4 were increased at the ADR-AU and ADR-SU stages.

SDFP-Growth Algorithm as a Novelty of Association Rule Mining Optimization

“…A rule with high confidence and a lift ratio equal to or greater than one is considered valid, but a rule with a lift ratio less than one, even with a high confidence value, cannot be considered valid. Equation 4shows the formula for calculating the lift ratio [33].…”